the choice of CPU and threads new cent kaby lake

if the processor has two cache on the same compute kernel can be so that the processor frequency above the ability to read and write to the cache. so a processor with 2 logical threads per core is able to read and write at the same time which can be equated to 2m action at a time.
accordingly, in practice, the system sees 2 cores regardless of whether they are physical or logical instead of 1 th. and can therefore have a speed tending to 2-fold when multiparallel task
actually of course you can think of any task where logical cores are slower than a real physical but rather tou difficult so that you can roughly equate Intel preprocessors as the core of =the flow and say as 4,8, 16 nuclear, etc.
about how I imagine this pitch will present so it can be harder still sealed.

and if someone knows what gives 14нм technology
pain in the ass pitch. the thing thou is that the more subtle the process is designed for lower voltage and frequency for corn need high voltage for chotko pounding. so in theory this chip is easier to damage both voltage and thermal deformation

simply these processors kaby lake 2 core 3.5+ ghz performance superior to fenopy 4 cores overclocking to 4 or more, etc.
if dryer 2 2009 the year that he had not pokazatelemvysokogo and sopostavit with the same Intel on 775 socket and some Coca i5 7хх already exceeds the dryer 2.
if not dryer 2 and older then the one that 3ядра 3 gig comp was finally he was already in his time insufficient to ilitny pitch gaming.
if you take today the Coca-U3 6100 or equivalent G I would characterize them as deficient and have no prospects of processors for office, home cement and lover of simple games.
options upgrades 1151 with the order that buying platform is with cheap CPU and then changes to the top I BYU not particularly considered mind dovolenka frequent changes combine and sockets from Intel. in terms of price to performance, say, year 4 will be more profitable kakya new platform with your Mat Board processor and memory. that's my Pitch of the story.
because let's say I now have some sort of gold school 2500к 2011 at 3770к as it is not very cool

MunchkiN 616
Thank you, neighing and infa is also useful

MunchkiN 616 wrote:
if the processor has two cache on the same compute kernel can be so that the processor frequency above the ability to read and write to the cache. so a processor with 2 logical threads per core is able to read and write at the same time which can be equated to 2m action at a time
He realized that he wrote? Processor frequency above the ability of the read/write cache O_o Clock frequency, roughly speaking, is how many cycles per second it can do in a unit time, and each operation requires n-th number of cycles, including read/write to the cache. How, then, do you have the clock frequency may be above the ability of the read/write cache? it's just a head explodes
MunchkiN 616 wrote:
and can therefore have a speed tending to 2-fold when multiparallel task
No, can't. High, how to write, normally written under multiphoton code, will tend to 50%
do not read further

fedr33 wrote:
people in the know explain what is 2 core and 4 threads compared to 4 threads 4ядра
Two cores four threads means that you have one material the core is divided into two virtual. For applications it is very bad, because the kernel system VirtualServer and divides its power into two, and these two cores are working with the app, not directly, but through a virtual environment. For processing large amounts of information in multitasking, it's the opposite is better, as it allows to spread the amount of information on several analysis centers. Therefore, in one application, the processor with no extra threads will run faster and with better performance.

fedr33 wrote:
simply these processors kaby lake 2 core 3.5+ ghz performance superior to fenopy 4 cores
AMD makes iron under upgrade not just. Current AMD processors at the time, a year or two they need to change if we want to keep high system performance. So AMD all the parts and are under constant upgrade.
Take old processors AMD makes no sense whatsoever. The Intel processors are more durable, they may be 10 years to work showing acceptable performance.

MunchkiN 616
But the author has asked about the difference of 2 cores with 4 threads and 4 cores with 4 threads.Me too this question interests.It's one thing when the core 2 flood.In the second case, how?So, in the first case, 6 core CPU.In the second-8.So?

[Denis]
Now I generally just try to explain. HT means simultaneous execution of two data streams on a single core. any CPU, executing the program produces a well-defined sequence of actions:
- load data from cache
- they are converted to understandable in the processor code. Performed by the decoder
- define the order of execution and actual execution
- combining and output the result
All these stages represent pipeline. So it turns out that I described the stage to perform so that all blocks of the CPU were running at full, does not work, why a part of resources idle. Including HT, all cores become virtual, the OS displays them separate kernel, but they are not full - physical core are considered major and have a higher priority in the execution of commands. Thus, the main core will always be to process commands, but if part of the resources of this core is idle, then they can go to another data flow that is connected to the additional core (virtual core physical). That's all. Accordingly, there is no about a 2-fold increase in performance can not out of the question, it will depend on how much resource busy the main thread (I mean physical core) and decided whether the developer to attend to that idle resources were loaded by another stream of data. In architectural terms, i.e. inside the processor, it looks like this:
- After the decoder decoded instructions (mops) are in the buffer, which features 2 stages, each of which is reserved for your thread. Accordingly, if suddenly all the resources are busy the main thread, then the idle can execute mops of the second stage (for example, Sandy and Ivy Bridge that have 2 turns at 28 mops)
- After decoder mops into the buffer, which offers a common place for main and additional (virtual) streams (for example, haswell General queue at 56 mops). So, in megapolice, with proper optimization, two threads can use a shared queue is more efficient, and kalapothaki one thread can use the entire queue, not just half as in the first case
Thus, 2 cores / 4 threads does not equal 4 cores / 4 threads. And twice there is no growth from virtual cores, a maximum of about 50% in sverhotvetstvennym applications, but usually 10-15 percent. Or maybe even become worse. Just from the technology score the more the CPU with fewer physical cores

[Denis]
this is where it all got caught on some nonsense -and in fact, no one wrote it )

Aktay
cool explained - thanks , even I understood )

that is going to sit on my phenom 965 and then nafig this Intel

fedr33
To understand, you need to change the percent or not, it is not enough just to know how it works HT. As I said, it all depends on the specific CPU and applications as a result of growth of HT may not be, may be minimal, and may be significant. But what is important is not only that, but the whole architecture as a whole in terms of how the blocks within the CPU and how, in principle, interact, and ability to follow instructions. Ie, any modern i3 faster than the hair dryer 965 th despite the 2 core/4 thread vs 4 full cores 965 th without regard to the question whether purchase i3)))

Aktay
I just needed to know what the flows and since the profit is not big reason to change no - Yes and i3 maybe faster but not enough to get under it a new mother and thought these might be steeper but no - I response to a question received - thank you !

AAA complex-complex-complex
finally have still the processor has a decoder and every translator. still have say so over the processor which deals with the fragmentation and the layout of megapolice that. and like how the app cannot be optimized under any CPU (and OS for all algorithms paralysis closed at the level of the processor architecture).
the essence of my pitch thought is that the block counters (I think a physical core) can perform numerical operations much faster than the whole transistor porridge with caches and translators. thus, relatively speaking, in the quantum adder bubbles appear when he did not fold and that means free time. NT, and other technologies of this kind of micro-parallelization of a denser help to use the mathematical apparatus of adders. naturally the ineffectiveness only when the frequency of the processor but not faster than running the task scheduler for this reason, and processors with more cores or some other technology sharing for example on NT 4 virtual flux and would be lost in performance.
the scarcity of my Pitch knowledge is ignorance whether stream code involving compute units of CPU or has its own logic
but in practice, the increase between the i5 and say 3570к 3770к does not exceed 20-30% in multiparallel benchmarks (((( and the difference between the I3 and 5 6ххх generation is about 50% in favor of these cores at the same frequencies.
so I was wrong with the evaluation of the effectiveness
then the problem probably is not the efficiency of flows and in the code itself, so a full parallel projected problems are almost nonexistent

Aktay
I have a rough idea what I wrote.But quickly forget.Thank you for the detailed explanation.If NT dubolomnyh technology,why it will ispolzuut?Isn't it better to leave a physical core?

fedr33
And I don't regret moving on with Panama rayzen,especially in monopolare:
Spoiler






Possible,with the old platform will be removed.
33 is a region?

[Denis]
Physical core mold harder and less profitable + it is not useless, just not all applications benefit from HT, but many even really need it. The dialect is the same, it all depends on the application and processor


MunchkiN 616 wrote:
finally have the same processor has a decoder and just translator
Decoder and every translator? How would a translator is an integral part of the decoder. More specifically, the decoder consists of a few simple translators, fewer complex and microsequencer with ROM. And when talking about TLB, it is the other step that is different from the decoder
MunchkiN 616 wrote:
still have say so over the processor which deals with the fragmentation and the layout of megapolice that
The plan involved the planners, the stage shall come into force after the decoder and dispatching. Moreover, the scheduler may not be, depending on the Manager type

MunchkiN 616 wrote:
the app cannot be optimized under some processor
Yeah, right, apps are not optimize for processors, I never did not claim. Just HT provides programmers with a scope for multithreaded programming, and the CPU will have to do what the programmer as it is taught to engineers doing


MunchkiN 616 wrote: the block adders (I consider him a physical core) can perform numerical operations much faster than the whole transistor porridge with caches and translators. thus, relatively speaking, in the quantum adder bubbles appear when he did not fold and that means free time
Complex representation. And what transistor porridge can be when all the blocks (caches, encoders, decoders, adders, registers, etc.) is a set of connected in a certain way transistors?! The metaphor of the adders and bubbles describes the situation where computing units are not fully what I wrote, in plain English, and not so in your atmosphere


MunchkiN 616 wrote:
naturally the ineffectiveness only when the frequency of the processor but not faster than running the task scheduler for this reason, and processors with more cores or some other technology sharing for example on NT 4 virtual flux and lost to speed
Yes, the job scheduler is limited, and a decoder, and controller, and speed of cache, and branch prediction... Yes anything it is limited))) the efficiency of ht depends mainly on how the program is written, and then all the other aspects

[Denis]
no it is not a region , you have rayzen 5 ? my budget is not that big at the moment ) bought the car - I just looked that all praise g4560 and g46 vhat else is there ( the same thing only higher GHz ) and I have even in the same DotA , GTA , shom - there are drawdown - and with each update the games - more and more ) so I was wondering how much more performance if I buy myself z170 and this g4xxx - the pocket not so much an effort .

Aktay wrote:
Physical core mold harder and less profitable
More precisely, there is no need. PC is not only and not so much of the game. PC development goes in the direction of multi-tasking in the Internet environment. Lots of running applications with a constant connection to the servers. For one application this is all redundant. Online games in contrast, processors with redundant threads perform better.

[Denis] wrote:
If NT dubolomnyh technology,why it will ispolzuut?
It's exactly the right technology, but the technology of Turbo boost dubolomnyh in which there is no need. It is made only for the sake of efficiency, which require legal documents. Processors without turbo boost application start immediately without thinking, so to work at full power immediately after starting. And processors with turbo buttom we have a little think, until they get on the operating frequency. In demanding games it is very noticeable. The same Mass Effect Andromeda launch, load save and you can go to smoke while all will raskoshelitsya. Then everything works as it should, no complaints, but here's a caveat of this technology.